model_info:
  name: anemll-Qwen-Qwen3-4B-ctx4096
  version: 0.3.5
  description: |
    Demonstarates running Qwen-Qwen3-4B on Apple Neural Engine
    Context length: 4096
    Batch size: 64
    Chunks: 2
  license: MIT
  author: Anemll
  framework: Core ML
  language: Python
  architecture: qwen3
  parameters:
    context_length: 4096
    batch_size: 64
    lut_embeddings: none
    lut_ffn: 4
    lut_lmhead: 6
    num_chunks: 2
    model_prefix: qwen
    embeddings: qwen_embeddings.mlmodelc
    lm_head: qwen_lm_head_lut6.mlmodelc
    ffn: qwen_FFN_PF_lut4_chunk_01of02.mlmodelc
    split_lm_head: 16
    vocab_size: 151936
    lm_head_chunk_sizes: [9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496]
    prefill_dynamic_slice: true

# =============================================================================
# Conversion Parameters (for troubleshooting)
# =============================================================================
# Generated: 2026-05-06 13:04:15
#
# model_path: /Users/optai/.cache/huggingface/hub/models--Qwen--Qwen3-4B/snapshots/1cfa9a7208912126459214e8b04321603b3df60c
# output_dir: /Users/optai/Documents/Anemll/converted_models/qwen3_4b_lut4_ctx4096
# command_line: ./anemll/utils/convert_model.sh --model /Users/optai/.cache/huggingface/hub/models--Qwen--Qwen3-4B/snapshots/1cfa9a7208912126459214e8b04321603b3df60c/ --output /Users/optai/Documents/Anemll/converted_models/qwen3_4b_lut4_ctx4096 --context 4096 --batch 64 --lut2 4 --lut3 6 --chunk 2
# context_length: 4096
# batch_size: 64
# num_chunks: 2
# lut_part1: none
# lut_part2: 4
# lut_part3: 6
# prefix: qwen
# architecture: qwen3
# argmax_in_model: false
# split_rotate: false
# single_cache: false
# dynamic_prefill_slice: true
# monolithic: false
# anemll_version: 0.3.5
# vocab_size: 151936
# lm_head_chunk_sizes: "[9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496, 9496]"
# =============================================================================